groundless-truth/CITATION.cff at main · releasecontrol/groundless-truth · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
cff-version: 1.2.0
message: "If you use this work, please cite it as below. Code and operational content are MIT-licensed; the six essays are CC BY 4.0. See LICENSING.md."
title: "Ground Truth Has No Ground: treating the label as a distribution in AI data labeling"
abstract: >-
  An argument and a runnable toolchain for AI data labeling. For contested,
  aesthetic, safety-relevant, and synthetic-human data there is no ground truth
  to recover, only a distribution of human judgement to preserve. The tools keep
  the label distribution, separate genuine human label variation from annotation
  error, flag value forks (cohort-level divergence) and manufactured consensus,
  and export trainer-ready soft labels plus a governance queue for decisions that
  majority vote would otherwise resolve silently. A JSON Schema specifies the
  resolution record each aggregation act produces: its inputs, rule, measures,
  authority, disposition, and replay hash. A decision-theoretic spine frames the
  label as a Bayes action -- an aggregation rule applied to evidence under a cost
  model -- decomposing it into four objects: the distribution, the decision, the
  rule, and the record.
type: software
authors:
  - family-names: Brandes
    given-names: Amadeus
    alias: AMBRA7592
repository-code: "https://github.com/releasecontrol/groundless-truth"
license: MIT
version: "1.4.0"
date-released: "2026-07-21"
doi: "10.5281/zenodo.21520646"
identifiers:
  - type: doi
    value: "10.5281/zenodo.21520646"
    description: "The version DOI for version 1.4.0 of this work."
  - type: doi
    value: "10.5281/zenodo.21520645"
    description: "The concept DOI for all archived versions of this work."
keywords:
  - data labeling
  - annotation
  - human label variation
  - inter-annotator disagreement
  - soft labels
  - pluralistic alignment
  - RLHF
  - data governance
  - information geometry
references:
  - type: article
    title: "The Problem of Human Label Variation: On Ground Truth in Data, Modeling and Evaluation"
    authors:
      - family-names: Plank
        given-names: Barbara
    year: 2022
    notes: "arXiv:2211.02570"
  - type: article
    title: "VariErr NLI: Separating Annotation Error from Human Label Variation"
    authors:
      - family-names: Weber-Genzel
        given-names: Leon
    year: 2024
    notes: "ACL 2024; arXiv:2403.01931"